AI Browser Automation Cost Analysis (Part 2): Proxy and Compute Optimization
Tokens first, proxy second, compute third. Proxy optimization doesn't save much money but reduces latency. Compute optimization matters at scale.
Engineering Blog
Tag: scaling
4 posts under this tag.
Chrome Headless at Scale (Part 3): Why K8s HPA Doesn't Work for Browser Clusters
CPU-based HPA is nearly useless for browser instances. Startup CPU spikes don't mean it's serving. Memory doesn't release after close(). Browsers aren't stateless services.
Chrome Headless at Scale (Part 2): CDP Connection Pool Exhaustion and Tab Accumulation
Tab 1 loads fine. Tabs 2 and 3 both time out. 120 seconds isn't enough. 1,600,000ms isn't enough either. The problem isn't page load speed — it's the connection pool.
Chrome Headless at Scale (Part 1): Orphan Processes and a 24GB Memory Leak
Unkillable Chrome processes, accumulating zombies, 24GB virtual memory leaks over 21 hours — scaling isn't just running 1,000 copies of the same instance.